Adventures with Image SensorsDr. Eric R. Fossum

Chairman and Chief Executive OfficerAdjunct Professor of Electrical Engineering, USC

2Mpixel Camera with Autofocus

Components

Axial Snubbers (2)

Lens Mount

Axial Snubbers (2)

Lens MountLens Barrel

Axial Snubbers (2)

MEMS Stage Assembly

Axial Snubbers (2)

MEMS Stage Assembly

Actuator Magnet Module

IR Filter

Actuator Magnet Module

IR Filter

EMI ShieldEMI Shield

Sensor & flex circuit

P l d S i C

Preload SpringHousing and Actuator Coils (2)

P l d S i C

Preload SpringHousing and Actuator Coils (2)

Preload Spring CapPreload Spring CapASIC driver

The Team

November 2005

Market Opportunity

• Camera phone market forecasted growth from over 600 millionCamera phone market forecasted growth from over 600 million units in 2006 approaching 900 million units by 2009

• Camera phone with resolutions of 2Mp+ projected to grow to more than 50% of total units during the same periodg p

• Cameras with 2Mp+ resolution desire Auto Focus capability for good image quality

Fossum’s Job

• CEO’s job is to turnSignificant market share

Additional customers2008

CEO s job is to turn this technology into a business that makes 2000K unit/month production

20072 d d t i d ti

Profitability

a profit for the investors.

500k unit production

2nd and 3rd customers

2006

2nd product in production

Product development

Preproduction

2005 Management team and Series C funding

Series B funding

Key customer engagement

Management team and Series C funding

EF joins Siimpel

Conceptual Design

First prototype2004

Business Thesis

• MEMS enables high precision integrated optical microsystems for high volume consumer products

Siimpel

• Handset applications:– Move to higher resolution (megapixel) camera

phones requires AF, zoom, shutters to produce acceptable image quality Siimpel

SiimpelMEMS device

acceptable image quality– Higher quality photo/video capability and

improved user experience of handset cameras increases customer usage and ROI for carriers

SiimpelAutoFocus

Module Design

– MEMS best meets handset requirements related to size, power, cost, performance and reliability

• Strategy is to sell or license high-margin camera module components such as MEMS AF actuator, MEMS shutter, MEMS zoon, ASIC driver, lens, etc. (generic phone)

MEMSCamera Phone Business Model

SILICON

SiimpelFocusCameraModule

HandsetBrand

MEMS & ASIC

PointOf

Sale

LicensedSiimpel

AssemblyStage

Fo ndr Flextronics MotorolaFlextronics

Flextronics Sony/EricssonFlextronics

EMS determines handset

SiimpelFoundry

NokiaFlextronics STM

Ericsson

Handset specifies EMS

NokiaSTMSTMHandset requires 2nd EMS source

MotorolaAltusAltusHandset requires 2nd EMS source

CONFIDENTIAL

Autofocus Actuator CompetitionVoicecoilMagnetic

StepperMotor

HelimorphPiezo-

electric

VariopticLiquid Lens

Konica-Minolta

Ultrasonic

SiimpelMEMS

Size Good Poor Poor OK OK Good

Cost Very Good OK OK OK OK OK

Optical Perform-

ance

OK Good OK OK low res OK Good

Power Poor Poor Good Good OK Good

Reliability Poor Unknown Very Poor Unknown Unknown Good

Scalable platform

Poor Poor OK Poor Poor Good

Precision/ Poor Good Poor OK low res Needs Excellent

Repeat-ability

feedback

CONFIDENTIAL

Competitive Barriers

• First to market with MEMS AF technology

AF d l d i l it i ll d d lti• AF module design complexity requires well rounded multi-disciplinary engineering team in critical areas of electro-magnetic, electro-optical and mechanical design which impact power, speed, shock specs

• Rapid pace of handset design changes and size and power constraints require ‘nimble’ engineering team and fast product development

• Building strong relationships with ‘Tier 1’ module and handset manufacturers

• Primary focus on high value MEMS device designs with y g gmodule component manufacturing outsourced to low cost manufacturers

• Strong IP and patent position

Some Of The Risks

• Miss insertion window due to series of “small” technical problems

• Large scale manufacturability not yet proven (t l / i ld > t)(tolerances/yield -> cost)

• AF camera phone market does not emerge• Competitors overcome their limitations faster/cheaper• Competitors overcome their limitations faster/cheaper

than expected

Business Development Status

• World’s largest camera module maker is investor and gfirst customer

• World’s 2nd largest camera module maker is enthusiastically evaluating technologyenthusiastically evaluating technology

• We are one of only 2 approved AF solutions for Motorola. Samsung to begin evaluationW f lfilli 500k it l d ti d• We are fulfilling a 500k unit early production order

• Working towards design in and design win for several handset opportunities

• Manufacturing capacity on ramp up from 10k units/week to 125k units/week to 500k units/week

Fossum’s Headaches

• Customer wants us to ramp to 2M-4M/month asap!p p• Capital Equipment $

– It costs about $10 in capex to increase capacity by 1 unit per month. (e.g., $10M for 1M/month capacity)( g , $ p y)

• Facility– It takes about 12,000 sq ft of space for 1M/month capacity

• People• People– It takes about 60 operators to build 1M/month– It takes a strong staff of engineers for manufacturing, new

product development and application supportproduct development, and application support• Where is a good 8” MEMS foundry when you need it??

How Did I Get Here?

• 1981-1983 Summer intern at Hughes Aircraft to work on i f d i

Adopted personal motto: “Have Fun, Make Money, Learn Along the Way”

infrared image sensors• 1984 Yale Ph.D. dissertation on smart image sensors• 1984-1990 Columbia EE Assoc Prof. on CCD focal plane p

image processing and GaAs CCDs• 1990-1996 NASA Jet Propulsion Laboratory - Imaging

Systems • 1995 Photobit spin off• 2001 Micron acquisition• 2003-2005 “Retirement”2003 2005 Retirement• 1999-present USC Adj. Professor• 2005-present Siimpel

• 16 Megapixel image sensors• 32 Megapixel UDTVg• What to do with gigapixel sensors

16 Megapixel sensor16 Megapixel sensor

From the Ph.D. dissertation research ofDr. Suat Ay

USC and Micron Technologygy

Big sensors for big questions

“FAST AND THE FURIOUS”

Credit: D. Wang (UMass) et al., CXC, NASASource : http://antwrp.gsfc.nasa.gov/apod/ap040906.html

Galaxy C153 plunges through galaxy formation ABELL 2125

A big chip

World’s largest array CMOS image Sensor

4096x4114

World’s largest single dieCMOS Integrated Circuit (IC)

and APS Image Sensor 4096x4114(16.85Mpixel)

and APS Image Sensor(58.98 cm2)

World’s largest well-depth imaging pixel

(1.35 Million electrons)( )

HPDPG APS Pixel Design

A' A'

PhotogatePhotodiode

A A

SelectPixout

RV

aa_pix

SFSEL

SelectR

SFSEL

PixoutV

aa_pixPhotodiode APS(REF2)

HPDPG APS

Reset

n+

RST

n+

ll ll

n+

BiasReset

Photogate

RST

n+n+

n well n well

n+

p- Epi Layer

n-well n-wellp- Epi Layer

n-well n-wellPhotodiode

Photodiode

Photodiode-Photogate Pixel

VAA

RST

CPIX

VAA

RSTVPG

Layout

Photodiode(PD)

Photogate(PG)VPG

CPIX

CPIXSEL

PG

PD

Light

V t RST=0

CPIX

SEL

OUT

C VPG=const. RST 0

n+ n+

VAACross

Section

• CPIX=CPD+CPG• C >>C and C ~C Photogate ON

p-Photogate Photodiode

• CPG>>CPD and CPIX~CPG Photogate ON• CPG<<CPD and CPIX~CPD Photogate OFF

PDPG Pixel Operation

VAA

RST=0

• Initial condition

Light CPIX

VPG=const.

n+ n+

p-

Strong Depletion • Reset the pixel• Start integrating photons• IntegratingPhotogate Photodiode

ΔV

VCH

GND

ΔVPIX

VAA

VRST

CPIXΔVPIX

V

ΔV

C2

C1

VSAT

ΔVPIX

Exposure(light*time)

PDPG Pixel Operation

VAA

RST=0

• Initial condition

Light CPIX

VPG=const.

n+ n+

p-

• Reset the pixel• Start integrating photons• Integrating• Reach “First Knee”

Photogate Photodiode

Strong Inversion

ΔVPIX

VCH

GND • Reach First Knee• Slow response

ΔVPIX

VAA

VRST

CPIXΔVPIX

V

ΔV

C2

C1

V1

VSAT

ΔVPIX

VT =VRST-VCHExposure(light*time)

E1

HPDPG Pixel – Light Sensitivity Measurement Results

5X Exposure enhancement

Decreased VPG

E1 E2

HPDPG Pixels in First Prototype Imager

Standard Photodiode

HPDPGPixels

Pixels

512x512 Array size, 15 μm pixel, 6-pixel design (PROTO1_2)

4Kx4K Device Architecture Partitioning

4Kx4K Imager after fabrication (before dicing)

6-inch Wafer

4Kx4K imager die

Controller and Demonstration System Solution

• Packaging PCB plugs onto the g g p gcontroller PCB,

• Controller does all the interface, ADC, DAC, timing, FIFO, bias

tigeneration.• Two ADC for channel readout,• Xilinx FPGA as main controller.• Analog frame grabber from

packaging PCB for analysis,

Reproduced 4Kx4K image

32 Megapixel UDTV32 Megapixel UDTV

From a collaboration betweenPhotobit Technology Corporation (now Micron)

AndNHK Laboratories, Tokyo

Ultra-Definition Television

8 Mpixel UDTV Sensor

8 Mpixel image (60 fps)

Chi Ph tChip Photo

UDTV Sensor PerformanceParameter Value Comments

Number of total pixels 3936 (H) × 2196 (V) Number of effective pixels 3840 (H) × 2160 (V) 8.3 M-pixels; 4 times the

HDTV resolution Pixel size 4.2 μm × 4.2 μm NW/P-sub photodiode APS

with on-chip microlensOptical format 1.25-inch Aspect ratio; 16:9 Die size 19.7 mm (H) × 19.1 mm (V) No stitching is used

0.06 LSB/electron Output referred Conversion gain 43 μV/electrons at pixel sense node

Full well capacity 25 000 electrons @ADC input range of 750 mVFull well capacity 25,000 electrons @ADC input range of 750 mVNoise floor 42 electrons Dynamic range 55.0 dB

4200 bits/lux-sec Output refereed 2700K light source, IR filter with cut-off wavelength of 650

Sensitivity

nm3.0 V/lux-sec at ADC input

Scanning Progressive Frame rate 60 fps Output signal 10-bit digital

16 parallel output ports

Output frequency 49.5 MHz 792 MHz (6.3 Gbps) total throughput

Electronic shutter Electronic rolling shutter Process technology 0.25 μm double-poly, triple

metal CMOS

Supply voltage 3 3VSupply voltage 3.3V Power consumption 597 mW at dark Package 262 pin PGA

Camera System

UDTV Camera

What to do with gigapixel sensors?What to do with gigapixel sensors?

Or how I spent part of my retirement

Sub-Diffraction Limit (SDL) Pixels

• Consider a pixel 0 5 microns x 0 5 micronsConsider a pixel 0.5 microns x 0.5 microns• A sensor of 24 mm x 12 mm is 48,000 x

24 000 pixels or 1 152 gigapixels24,000 pixels or 1.152 gigapixels.4 um

Airy Disk

0.5 um

Airy Disk3.7 um

At 550 nm

Same F#, Scene, Exposure

4 um pixel

0.5 um pixel

5000 e-/exposureShot noise SNR = 71:1Capacity 20,000 e-

78 e-/exposureShot noise SNR = 8.8:1Capacity 312 e-p y

5 e- read noiseDynamic range = 72 dB

p y5 e- read noiseDynamic range = 36 dB

SDL Pixels Hardly Make Sense

• Can’t collect much light or chargeCan t collect much light or charge • Can’t hold much charge

C ’t i d t il t i l l ti• Can’t image details at pixel resolution• Uniformity questionable

What to do with SDL Pixels?

• Oversampling of spatial and colorOversampling of spatial and color information?

Digital Film• Each pixel could act like a binary detector (photon hit or not) – call it a “jot”• Jot doesn’t need any capacity or much uniformity• Trade off grain size (resolution) and “film speed”• D-Log H exposure behavior

Hit jots Develop 4x4 grains

1 gigajots 72M binary pixels ~1M pixel

Making Nano-scale Pixels

• Pixels need not be silicon PN-junction-like – justPixels need not be silicon PN junction like just need to be able to “flip” if hit by photon

• Quantum dots?Q• Molecular pixels?• a-Si:H or organic pixels?a Si:H or organic pixels?• Probably still need silicon readout.